Replace accidental O(N^2) BFRT with O(N log N). 13% better on NETLIB

[chris-maes]

The current bound-flipping ratio test (BFRT) makes an O(N) pass through the remaining breakpoints for each breakpoint resulting in O(N^2) run time, where N is the number of breakpoints.

This PR places the breakpoints in a heap in O(N) time. And then removes the minimum breakpoint O(1) and reorders the heap O(log N) for each of the N breakpoints. Thus, the run time should be O(N log N).

On the NETLIB LP test set this results in a 67% reduction of time spent in the BFRT and a 13% reduction in overall solve time.

The entering variable chosen by the new code should be the same as the previous code (except when ties are broken differently when constructing the heap).

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[chris-maes]

This is still a work in progress. As I have not yet removed the old code, since I am still using it to run benchmarks on the MIPLIB test set.